sails

Adding from Mac perspective, I am also keeping an eye on Linux. I’ve hit a wall with Mac window management, and find the operating system just gets in the way for professional use across multiple of their digital “desktops”. I have no useful way to isolate work streams, and would gladly move to something better.

The blocker for Linux for me as someone who wants some level of reliability has always been fiddling with low level config, but now with Claude Code, low level config appeals!

Cool thanks, yes all good, yes there is basically a combo of knowledge dump and assembling raw weather data to apply the knowledge to, I haven’t done another location yet, but maybe wait for a few more requests :D

I had lots of fun coercing smaller local models into using tools.

I found that designing the tools (ergonomics) mattered much more than designing the system using the tools.

Nice, does require some local knowledge to get it at all useful (if you can email me some info in what conditions work I can set one up for you.)

Is there a high res weather model for Aus? That makes a huge difference as well

I’m doing a similar thing with open-meteo for surf forecasts (for myself primarily)

Only one region, but could quite easily expand. It takes the open-meteo ocean data and combines it with some short and long range weather. Then run a preprocessed refined version of that through an LLM to turn it from quantitative into qualitative. It basically does what I would do in my head.

If you have any ideas, please let me know

https://surfrash.xyz/

I like this for adjacent things too.

Company website in the same repo means you can find branding material and company tone from blogs, meaning you can generate customer slides, video demos

Going further, Docs + Code, why not also store Bugs, Issues etc. I wonder

I've tried these in the past but always found the software/firmware side of it too fiddly for the reward (against the grain for a lot of people, I know), which is a pity as the hardware side is really rewarding (for me).

Recently, doing it with Claude Code was a breeze. If you are more interested in the outcome than the process, then I'd say it's a great time to buy a few old Kindles and see what you can create with them.

Also have a look at cube, I saw some similarity, I much prefer your clean approach https://cube.dev/docs/product/data-modeling/overview

This is really interesting, as a data analyst and product person I like this embedded data modelling approach.

I would be interested in knowing if you’ve thought about what to do with external data (eg Stripe) and data processing engines (eg Snowflake/Databricks/clickhouse or duckdb)

These are great. Something you might find interesting is that you can expose a google sheet to have an interactive database. I have a map similar to yours, but with surf spots. Maybe defeats the point, but I find it handy

Edit: come to think of it, I should revisit it now that everyone can vibe code. The sheet was to allow people to add to it, now maybe easier for me to take a message and ask an agent to update the html directly

Posted a video in the thread, it’s pretty rudimentary (Claude code does everything) at the moment but I think this has a lot of possibilities.

Mostly using it for technical marketing/explainer videos eg https://x.com/mattarderne/status/1987441582413345016

I’ve had a lot of fun with Remotion and Claude Code for CLI video editing. I’ve been impressed with how much traditional video editing I can manage.

I will be checking this out!

As others have explained, ensembles are useful.

As a layperson, what _is_ useful is to look at the difference between models. My long range favourite is to compare ECMWF and GFS27 and if the deviation is high (windy app has this) then you can bet that at least one of them is likely wrong

Agree, it feels so fundamental. Any idea why? Gemini has also had it for a long time

Good guide.

In the spirit of finding the right person through serendipity:

I’m actively looking for intros to angels who are interested in B2B lending. The tariffs have completely disrupted the US SMB lending ecosystem and automation is going to have to ramp up rapidly to close the gap.

Myself and my team have a AI automation tool, we are revenue generating, technical founding team with 2x maths PhDs

Please email me (email in profile) if relevant to someone you know

$5 could go towards a strict AI reject/review funnel as a prefilter

Agree, each agent creating a PR and then coordinating merges is a pain.

I’d like

- agent to consolidate simple non-conflicting PRs

- faster previews and CI tests (Render currently)

- detect and suggest solutions for merge conflicts

Codex web doesn’t update the PR which is also something to change, maybe a setting, but for web Code agents (?) I’d like the PR once opened to stay open

Also PRs need an overhaul in general. I create lots of speculative agents, if I like the solution I merge, leading to lots of PRs

I’m amazed they published it with such a poorly applied analogy.

I feel like we are already there. I would imagine if you set Claude Code or Codex this task, running in the CLI, you would see a huge improvement, and that is before you start creating task specific guardrails.

I’m surprised they haven’t tried this, I’m running my own in parallel against my accountant in this way.

Yes I thought so too.

I wonder if it will mean more or less revealing of the models that are running agentic flows (we currently abstract with Fast/Smart)

It is also possible that the first model calls other models, and you could reverse engineer the tool call structure by seeing when glitches occur based on different branches of the tool calling

> This isn't just a coding problem; it's a reasoning problem.

I’ve been trying to get Claude to make 3d visuals for Remotion.dev

It does a reasonable job, but really struggles to do more complex 3D visualisations.

It feels like a lack of feedback problem. It can’t validate when the output is wrong (like frontend design I guess).

Interested to know what models generated the code

I’d be interested in testing different data formats when using the structured outputs api

This article is about enshittification not Amazon

> Nvidia’s vendor financing becomes exposure to customers building competitive alternatives.

This is surely the most important line in the piece? In what world would this much demand not lead to alternatives emerging?

(Assuming the upside, yes if the demand is not there in two years then yes it’s all going to burn)

Thanks for the feedback, unfortunately we need some auth, this covered the most bases.

> The key insight: when you interact with modern ChatGPT, you’re not just talking to a language model - you’re collaborating with an intelligent agent that uses language models as one component of a much more sophisticated cognitive architecture that includes memory, tools, and orchestration.

Fwiw, right at the end

Simon is one of the few people exploring the contemporary challenges of data modelling in a meaningful way.

I would LOVE to have this as my physical desk top.